The millimeter-wave hardware I build

The first thing I learned about millimeter-wave hardware is that the interesting part is never the part on the slide. Above 30 GHz the chip, the board, the connector and the antenna stop being separate problems with tidy boundaries. A transition a millimeter too long turns into a resonator. A solder bump turns into an inductor. The package that was supposed to protect the chip starts radiating. Whether a design survives contact with the real world is decided in those joints, and the joints are where I have done my best work.

It began with one of the least forgiving joints there is: getting a signal out of a standard rectangular waveguide and into an air-filled substrate integrated waveguide, a channel of air carved inside an ordinary multilayer board. Air is the best dielectric there is, which is why the structure matters at high frequency. But the accepted way to design the transition into it was to guess a shape, simulate it, nudge it, and repeat until the plot looked acceptable. Fifty runs was normal. I wanted the first run to be right.

So we worked the fields at the junction through by hand and wrote design equations that give the geometry directly from the band and the two cross-sections. Then came the part that separates a result from an invention. The machined prototypes came back and did not match the screen. A channel a few microns narrow, a bonding layer thicker than the datasheet, a flange that was not quite flat. Weeks of chasing each one. The equations held in the end, and they held because the structure had been chosen from the start to tolerate exactly those errors. The measured transition covered the whole WR-15 band, 50 to 75 GHz, with less than a quarter of a decibel lost on the way through, and the measurements sat on top of the simulated curves. The measured parts went to the IEEE International Microwave Symposium, the method went to IEEE Access, the apparatus went to the US Patent Office and came back as two patents, and the work took a Gold Medal at the International Exhibition of Inventions Geneva, among more than a thousand inventions from over forty countries.

From there the work widened: multi-beam and tilted-beam air-filled horn arrays for satellites and urban air mobility, in rigid and flexible forms, covering the entire V-band at 15.5 dBi with four independently tilted beams and a fold that lets the array pack flat for launch; a 250 to 350 GHz slot antenna in SiGe BiCMOS; a review of every practical way to connect SIW circuits to the rest of a system; and, through an IEEE Electronics Packaging Society fellowship with Fraunhofer HHI, a proposal for 6G antennas designed together with their package instead of in spite of it. Most recently, a low-cost dual-polarized magnetoelectric dipole for early 6G distributed systems: 44 percent bandwidth, more than 40 dB of isolation between its two polarizations, 7 to 8 dBi of gain, and a body that is 3D-printed plastic with electroplated copper so that a distributed network could afford thousands of them.

The projects look different from the outside and are the same problem underneath. A satellite array, a drone antenna and a terahertz slot on a chip all live or die in their joints: where the signal leaves one medium for another, where the board meets the package, where the antenna meets the world. The habit the first transition taught, of writing the physics down before touching a solver and of choosing the structure that forgives the shop its errors, is the habit that carried every project after it. It is slower on day one and faster by the first prototype, and it is the reason the parts that come back from fabrication tend to do what the drawing said.

Two of those projects point at where the field is going. The 250 to 350 GHz slot antenna is an experiment in putting the antenna on the same piece of silicon as the transistor, so that the most fragile connection at those frequencies stops existing. The flexible arrays are an answer to the platforms that 6G is meant to serve, aircraft skins, vehicle bodies, deployable panels, none of which is flat. Both are cases where the packaging is the design, not a step after it, and both are why I spent a fellowship year proposing 6G antennas that are conceived together with their package rather than fitted into one later.

What comes next is the joint between disciplines rather than materials. The most expensive mistakes of the coming decade will not be made inside a chip or inside an antenna. They will be made in the assumptions each team makes about the other: the array that cannot be cooled, the package that detunes what it protects, the radio that works on the bench and not in the housing. The bench taught me how to close the distance between a simulation and a working part. The company I am building applies the same discipline to the distance between hardware, chips and AI. The essays on this site are the notes along the way.

Photos: NOAA National Severe Storms Laboratory.